Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 27 Mar 2002 16:51:45 -0500 (EST)
From:      John Baldwin <jhb@FreeBSD.org>
To:        Alfred Perlstein <alfred@FreeBSD.org>
Cc:        current@FreeBSD.org, Kris Kennaway <kris@obsecurity.org>
Subject:   Re: Lock order reversals in sys_pipe.c
Message-ID:  <XFMail.20020327165145.jhb@FreeBSD.org>
In-Reply-To: <20020326224745.GF93885@elvis.mu.org>

next in thread | previous in thread | raw e-mail | index | archive | help

On 26-Mar-2002 Alfred Perlstein wrote:
> * Alfred Perlstein <alfred@FreeBSD.org> [020326 14:43] wrote:
>> * Kris Kennaway <kris@obsecurity.org> [020324 14:26] wrote:
>> > The bento cluster is now running with WITNESS enabled to try and track
>> > down some odd UMA lock corruption panics.  Instead, it found the
>> > following lock order reversal in sys_pipe.c overnight:
>> > 
>> > Mar 24 09:02:29 <user.crit> gohan13 kernel: lock order reversal
>> > Mar 24 09:02:29 <user.crit> gohan13 kernel: 1st 0xd99d6500 pipe mutex @
>> > /local0/scratch/usr/src/sys/kern/sys_pipe.c:450
>> > Mar 24 09:02:29 <user.crit> gohan13 kernel: 2nd 0xd971cddc process lock @
>> > /local0/scratch/usr/src/sys/kern/kern_sig.c:2093
>> > 
>> > Those source references are from a -current kernel from last night.
>> 
>> Are you %100 on that?  How did you get this to happen?
>> 
>> I can see where I hold the pipe lock, then try to get a proc lock,
>> but not the other way around...
>> 
>> Any ideas?
> 
> Note that I can pretty trivially fix this by dropping the pipe lock
> while calling pgsigio in pipeselwakeup(), however I will then have to
> make sure the callers of pipeselwakeup() can deal with someone else
> mucking with the pipe during the call.
> 
> Anyhow, wtf doesn't witness print out at least one instance of the
> old locking?  Basically let me know where locking is happening in
> the other direction?

That's easy to achieve.  Add a static lock ordering in the compiled in arrays,
then it will let you know when you do the other way.  It may be an indirect
relationship and not a direct one as well.  The allproc/filedesc thing was an
instance of indirect relationships determining the order.

-- 

John Baldwin <jhb@FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve!"  -  http://www.FreeBSD.org/

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?XFMail.20020327165145.jhb>